ADAPTIVE BUILDING EDGE DETECTION BY COMBINING LIDAR DATA AND 
AERIAL IMAGES 
LI Yong, WU Huayi 
State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, 
Wuhan 430079, China - liyongwhu@gmail.com, wuhuayi@lmars.whu.edu.cn 
KEY WORDS: LIDAR, Aerial image, Fusion, Edge detection, Building extraction, DEM 
ABSTRACT: 
The building edge detection plays a key role during building extraction, which is important and necessary for building description. 
The edges detected from aerial images have high horizontal accuracy and represent various edge shapes well. But the edge detection 
in images is often influenced by contrast, illumination and occlusion. LIDAR data are suitable forjudging building regions, but miss 
some edge points due to the laser pulse discontinuousness. In order to make full use of the complementary advantages of the two 
data sources, a new adaptive method of building edge detection by combining LIDAR data and aerial images is proposed in this 
paper. Firstly, the objects and ground are separated by a filter based on morphological gradient. The non-building objects are 
removed by mathematical morphology and region growing. Secondly, the aerial image is smoothed by Gaussian convolution, and 
the gradients of the image are calculated. Finally, the edge buffer areas are created in image space by the edge points of the 
individual roof patch. The pixels with local maximal gradient in the buffer area are judged as the candidate edge. The ultimate edges 
are determined through fusing the edges in image and the roof patch by morphological operation. The experimental results show that 
the method is adaptive for various building shapes. The ultimate edges are closed and thin with one-pixel width, which are very 
suitable for subsequent building modelling. 
1. INTRODUCTION 
1.1 Background 
Building models are the essential components of the three 
dimensional GIS. The building edges are the significant features 
of buildings. The building edge detection plays a key role 
during building extraction, which is important and necessary for 
building description. The aerial images and LIDAR data all 
provide powerful support to the task. The edges detected from 
aerial images have high horizontal accuracy and represent 
various edge shapes well. LIDAR point clouds are well suited 
for judging the points that belong to each building surface, 
which is beneficial to search the approximate location of each 
building edges. However, each data source has its own 
weakness beside strength for the edge detection. 
In aerial images, the contrast between buildings and 
backgrounds is often not so high, and there are too many 
complex spectral and texture information in most of scenes, 
including occlusion, shadows and so on. Those lead to the 
complexity of edge detection. Some edges are not building 
edges we need, while some edges of buildings are missed or 
broken. Furthermore, edges of different objects or edges of 
different layers of one building are likely to stick to each other. 
In LIDAR point clouds, building edges can be located by 
analyzing the height changing of laser footprints. However, 
some edge points are not gathered by LIDAR for the laser pulse 
discontinuousness, which cause that the horizontal accuracy of 
edge detection from LIDAR data is poor. 
1.2 Related work 
As discussed above, both data sources have their own pros and 
cons. In order to overcome the drawbacks using single data 
source, it is considered as a promising strategy to automatically 
detect building edges by fusion of LIDAR data and aerial 
images. The correct edge detection is a challenging task due to 
the scene complexity. So it is a research hot spot how to 
combine the two different data sources in an optimal way so 
that their weakness can be compensated effectively by each 
other. 
(Rottensteiner and Jansa, 2002) firstly generate initial 3D 
planar segments of buildings from LIDAR point clouds, and 
create polyhedral models and wire frames of buildings by 
analyzing relations of neighbouring planar segments and 
regularizing of building shape. Then, the initial polyhedral 
building models are verified in the images to improve the 
accuracy of their geometric parameters. The wire frames of 
buildings are back-projected to the images to match with image 
edges for improving the accuracy of the building outlines. Only 
straight line segments are matched with the object edges. 
(Sohn and Dowman, 2007) collect rectilinear lines around 
building outlines by data-driven and model-driven ways for 
building modelling. In the data-driven way, straight lines that 
are significantly long and around building boundaries are 
extracted from optical imagery. Then the lines are 
geometrically regularized by analyzing the dominant line angles. 
The lines produced in the data-driven way do not always cover 
all parts of building edges because significant boundary lines 
may be missed due to low contrast, shadow overcast and 
occlusion effects. In the model-driven way, new lines are 
extracted from point clouds in order to compensate for the lack 
of data-driven line density by employing specific building 
models based on the assumption that building outlines are 
comprised of parallel lines. So only the building edges with 
parallel and orthogonal structure can be considered. 
197